Control of biped locomotion inspired from inspired from walking in - - PowerPoint PPT Presentation

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Control of biped locomotion inspired from walking in monkeys

Charles-Henry Houdemer

Biorob Ecole polytechnique fédérale de Lausanne LSRO École polytechnique fédérale de Lausanne

18 June 2010 / Final Presentation

Supervisors : Sarah Dégallier, Jesse Van den Kieboom, Solaiman Shokur, Auke Jan Ijspeert

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Outline

1

Introduction Insights on BMIs Previous work Goal of the project Challenges Long term goals

2

Results Webots model Numerical tests CPG model Results

3

Future improvements

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Insights on BMIs

Three different forms :

Invasive Partially-invasive Non-invasive

Two types :

Motor BMI Sensory BMI

Transform data either from or to the brain by predicting and interpreting correct behaviour

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Insights on BMIs

Three different forms :

Invasive Partially-invasive Non-invasive

Two types :

Motor BMI Sensory BMI

Transform data either from or to the brain by predicting and interpreting correct behaviour

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Insights on BMIs

Three different forms :

Invasive Partially-invasive Non-invasive

Two types :

Motor BMI Sensory BMI

Transform data either from or to the brain by predicting and interpreting correct behaviour

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Insights on BMIs

Three different forms :

Invasive Partially-invasive Non-invasive

Two types :

Motor BMI Sensory BMI

Transform data either from or to the brain by predicting and interpreting correct behaviour

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Insights on BMIs

Three different forms :

Invasive Partially-invasive Non-invasive

Two types :

Motor BMI Sensory BMI

Transform data either from or to the brain by predicting and interpreting correct behaviour

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Previous work

Fitzsimmons’ work on monkeys1 :

Model try to predict the position of each joint Quality of predictions is good for one direction but degrades for two directions Must switch between two models for forward and backward

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Experimental setup

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Achievements

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Goal of the project

Build a CPG with the data from Nicolelis Lab in Duke University Data consists in Cartesian positions of hip, knee and ankle positions, speed of walking and indication if the foot is on the ground Reproduce the walking gait of a rhesus monkey with this CPG recieving basic inputs from a BMI Generalize it if possible to :

Different sessions; Different speeds; Different monkeys; Both directions.

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Goal of the project

Build a CPG with the data from Nicolelis Lab in Duke University Data consists in Cartesian positions of hip, knee and ankle positions, speed of walking and indication if the foot is on the ground Reproduce the walking gait of a rhesus monkey with this CPG recieving basic inputs from a BMI Generalize it if possible to :

Different sessions; Different speeds; Different monkeys; Both directions.

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Goal of the project

Build a CPG with the data from Nicolelis Lab in Duke University Data consists in Cartesian positions of hip, knee and ankle positions, speed of walking and indication if the foot is on the ground Reproduce the walking gait of a rhesus monkey with this CPG recieving basic inputs from a BMI Generalize it if possible to :

Different sessions; Different speeds; Different monkeys; Both directions.

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Goal of the project

Build a CPG with the data from Nicolelis Lab in Duke University Data consists in Cartesian positions of hip, knee and ankle positions, speed of walking and indication if the foot is on the ground Reproduce the walking gait of a rhesus monkey with this CPG recieving basic inputs from a BMI Generalize it if possible to :

Different sessions; Different speeds; Different monkeys; Both directions.

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Challenges

Monkeys are not naturally bipedal Monkeys’ behaviour is not always predictable

Play with holding bar Make feet slide on the ground Video Causes variability in measurements

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Typical variations in the hip angles

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Long term goals

Allow paraplegics to regain the control of their legs Allow for more robust predictions

Less informations are needed Less neurons are needed May lose more neurons before losing efficency

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Long term goals

Allow paraplegics to regain the control of their legs Allow for more robust predictions

Less informations are needed Less neurons are needed May lose more neurons before losing efficency

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Long term goals

Allow paraplegics to regain the control of their legs Allow for more robust predictions

Less informations are needed Less neurons are needed May lose more neurons before losing efficency

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Creation

Built from the webots human model of Jesse van den Kieboom Only hand size and total weight were known Size and masse extrapolated with data from Hamada2 Main differences :

Tail added; Head lighter than with normal division to reflect that a monkey head is lighter than a human one.

Uses a grannywalker to prevent it from falling on its side

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Monkey and human models

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Curves were normalized for testing

CPGs can be easily scaled but not change shape

For each cycle :

Amplitude was normalized ˆ y = (y − ¯ y)/(ymax − ymin) Frequency was normalized by scaling each time interval to a reference one

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Duty factor

Duty factor

Decreases with speed Right leg different than left one Variance lower with high speeds

Equation of the form (for one cycle) :

stancetime totaltime

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Duty factor

Duty factor

Decreases with speed Right leg different than left one Variance lower with high speeds

Equation of the form (for one cycle) :

stancetime totaltime

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Duty factor

Duty factor

Decreases with speed Right leg different than left one Variance lower with high speeds

Equation of the form (for one cycle) :

stancetime totaltime

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Duty factor

Duty factor

Decreases with speed Right leg different than left one Variance lower with high speeds

Equation of the form (for one cycle) :

stancetime totaltime

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Square errors

Square errors

Steady step Mean error low Variance relatively low

Equation of the form (for one cycle) :

n

i=1(yiknown − yipredicted)2

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Square errors

Square errors

Steady step Mean error low Variance relatively low

Equation of the form (for one cycle) :

n

i=1(yiknown − yipredicted)2

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Square errors

Square errors

Steady step Mean error low Variance relatively low

Equation of the form (for one cycle) :

n

i=1(yiknown − yipredicted)2

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Square errors

Square errors

Steady step Mean error low Variance relatively low

Equation of the form (for one cycle) :

n

i=1(yiknown − yipredicted)2

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Square errors of Nectarine - Hip

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Square errors of Nectarine - Knee

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Design of the CPGs

Data were in Cartesian form and contained the positions of the knee, hip and ankle CPG is of the form :

˙ xi = γi(fi(θi) − xi) + dfi

dθi · ˙

θi + Ki

Phase coupling is of the form :

p = sin(from.p ∗ 2 ∗ π − to.p ∗ 2 ∗ π − bias)

A cycle was arbitrarily chosen for the building of the CPG

Average Starts and ends approximatively at the same height

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Design of the CPG

Joint angle cycle was cut in three parts for interpolation Parts were interpolated with third degree polynomials Polynomials were used with cpgstudio Testing on webots was done with the cpg file obtained through cpgstudio and the libcpg, both from Jesse van den Kieboom The necessary data for building a CPG for the ankle and adapting the foot were obtained by tracking positions on a video with the help of Kostas

• Karakasiliotis. The implementation of the CPG was

abandoned in order not to add more complexity to an already well performing model.

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

CPG output

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Real and interpolated curves - Hip

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Real and interpolated curves - Knee

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Visual tests

Can walk forward and backward Handles different speeds

Limited by ground adherence

Video

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

Future improvements

Implement flexible feet for more efficient foot contact Modify the polynomials to reduce the error in interpolation further in order to obtain a interpolated curve even more close to a real one

Control of biped locomotion inspired from walking in monkeys Charles-Henry Houdemer Introduction

Insights on BMIs Previous work Goal of the project Challenges Long term goals

Results

Webots model Numerical tests CPG model Results

Future improvements References

References

1

Nathan A. Fitzsimmons, Mikhail A. Lebedev, Ian D. Peikon,and Miguel A. L. Nicolelis. Extracting kinematic parameters for monkey bipedal walking from cortical neuronal ensemble activity.

2

Yuzuru Hamada, Nontakorn Urasopon, Islamul Hadi, and Suchinda

• Malaivijitnond. Body Size and Proportions and Pelage Color of

Free-Ranging Macaca mulatta from a Zone of Hybridization in Northeastern Thailand.